Cooperation, Correlation and Competition in Ergodic $N$-player Games and Mean-field Games of Singular Controls: A Case Study

Abstract

We consider ergodic symmetric $N$-player and mean-field games of singular control in both cooperative and competitive settings. The state process dynamics of a representative player follow geometric Brownian motion, controlled additively through a nondecreasing process. Agents aim to maximize a long-time average reward functional with instantaneous profit of power type. The game shows strategic complementarities, in that the marginal profit function is increasing with respect to the dynamic average of the states of the other players, when $N<\infty$, or with respect to the stationary mean of the players' distribution, in the mean-field case. In the mean-field formulation, we explicitly construct the solution to the mean-field control problem associated with central planner optimization, as well as Nash and coarse correlated equilibria (with singular and regular recommendations). Among our findings, we show that coarse correlated equilibria may exist even when Nash equilibria do not. Additionally, we show that a coarse correlated equilibrium with a regular (absolutely continuous) recommendation can outperform a Nash equilibrium where the equilibrium policy is of reflecting type (thus singularly continuous). Furthermore, we prove that the constructed mean-field control and mean-field equilibria can approximate the cooperative and competitive equilibria, respectively, in the corresponding game with $N$ players when $N$ is sufficiently large. To the best of our knowledge, this paper is the first to characterize coarse correlated equilibria, construct the explicit solution to an ergodic mean-field control problem, and provide approximation results for the related $N$-player game in the context of singular control games.